We start and buy companies using thrvAI and our Jobs-to-be-Done method.
We find complex customer jobs in large markets and build new agentic intelligence products with our thrvAI equity value creation platform. Once we have product-market fit, we go to market by starting a new company or acquiring an existing company with the target customers. If you are a limited partner, a VC investor, a PE sponsor, or a corporate venture investor contact us to learn more.
We use thrvAI to analyze hundreds of markets. The markets that meet our criteria become new products for a start up or an acquisition.
- Markets100s analyzed
- thrvAI builds v1New agentic product
- Product-market fitIterated with agents
- Go-to-marketStart-up or Acquisition
Customers are not buying a product. They are hiring a product to get a job done.
Clay Christensen
Who we are
We have refined our method since 2013.
Now we use it to
start and buy companies.
A method built with leading companies.
We developed our Jobs-to-be-Done method and platform with Microsoft, Google, American Express, eBay, Oracle, Twitter and Target. Our method identifies the unmet needs in the customer’s job and the weaknesses of competitors, based on how fast and accurately their products help the customer get their job done.
A track record of creating equity value.
We have used our method to create equity value with the companies we work with. You can read the case studies and hear directly from the CEOs we have worked with.
Experienced operators and investors.
Jay Haynes, our founder and CEO, was an investment professional at GTCR, a product manager on Windows at Microsoft, and CEO of two startups, one funded by Kleiner Perkins and the other by DFJ. Jay leads the thrv team of executives from leading technology companies, supported by experienced advisors. Learn more about us.
- v1SaaS era
2013 - v2SaaS era
2015 - v3SaaS era
2020 - thrvAIAgentic Platform
2023 - thrv PortfolioStarting and acquiring companies
2026
“The application layer on top, ultimately, is where economic benefit will happen.”
Jensen Huang
Our method
Technology changes.
The customer’s job does not.
This is a competitive advantage for our portfolio companies.
Theodore Levitt of Harvard Business School famously said, “Customers don’t want a quarter-inch drill, they want a quarter-inch hole.” This is true in every market. Products change with each wave of technology, from the CPU in the 1970s to the internet in the 1990s to AI in the 2020s, but the customer’s job stays the same. This is why the customer’s job is the North Star and competitive advantage for our portfolio companies. The following are three examples where the market can be redefined using the customer's job:
Consumer
Consumers don’t want iPods.
They want to create a mood with music.
- Vinyl 1964
- Optical 1984
- Hard drive 2001
- Cellular 2007
- Streaming 2021
Business
Salespeople don’t want CRM.
They want to acquire new customers.
- Phone 1876
- Email 1972
- On-premise 1993
- SaaS 2002
Health
Surgeons don’t want stents.
They want to restore artery blood flow.
- Amputation 1800
- Balloon 1964
- Stent 1986
- Drug-eluting stent 2002
The customer’s job
Every job has traffic jams.
Our portfolio companies find the alternative routes.
Customer jobs are complex. A typical job has 15 to 20 steps to plan, execute, assess, revise and conclude the job. Each step has 10 to 15 customer needs, for 120 to 150 needs in total, and every job has metrics, root causes, constraints and emotions.
Take the job of getting to a destination on time. Like every job, it has the equivalent of traffic jams that keep the customer from getting the job done. The customer has to make important and difficult decisions about whether to take an alternate route to arrive on time.
With thrvAI, we build new products for our portfolio companies that are the equivalent of full self-driving agents: intelligent agents that get the job done for the customer as automatically as possible to create economic value for the customer.
- 20Job steps
- 140Needs
- 25Metrics
- 35Root causes
- 40Constraints
Get to a destination on time
- Determine the routePlan
- Travel to the destinationExecute
- Assess if the arrival time is accurateAssess
- Revise the route if neededRevise
- Arrive at the destinationConclude
Companies that survive disruption build new products with autonomous disruption teams that build new and independent businesses around the disruptive technology.
Clay Christensen
From job to strategy
The job is the market.
thrvAI finds the strategy for our portfolio companies.
We define each market for each of our portfolio companies by the customer’s job, not by the products. Customers want holes, not drills. The customer’s job is the market.
We pioneered and patented sizing market opportunities by the customer’s willingness to pay to get the job done rather than by the price of current products. This is how we size the markets for our portfolio companies. This same analysis is how we identify the target customer segment and the ideal customer profile for our portfolio companies.
thrvAI identifies the unmet needs in each step of the job and where competitors’ products fall short in getting the job done fast and accurately.
thrvAI uses this analysis to set the optimal strategy for an agentic intelligence product that gets the job done and creates economic value for customers.
Our Investment criteria
What we look for when we start or buy a portfolio company.
With thrvAI, we are analyzing hundreds of business, consumer and health markets against five investment criteria.
Our first criterion is a large market with a complex customer job, where customers struggle to get the job done quickly and accurately with the competitive products in the market today.
We look for jobs where it is difficult for customers to make the important decisions that determine whether they succeed.
We look for jobs where intelligence and compute can get the job done faster and more accurately.
We target large markets, sized by the customer’s willingness to pay to get the job done, the frequency of the job execution, the number of customers in the market, the economic value of the job, and the opportunity to disrupt the incumbents.
We look for jobs where getting the job done better has a meaningful impact on people’s lives. This is the criterion we weigh most.
Disruption
Agents are a disruptive technology for software.
Clay Christensen showed that incumbents improve their products along a sustaining trajectory for their most demanding customers. Eventually they overshoot what most customers can use. A disruptive technology enters below mainstream needs, improves faster, and takes over the market with a different business model.
Legacy software is on the sustaining trajectory. It keeps adding features that organize the inputs to the customer’s job, while the customer still does the work. Agentic intelligence products do the job for the customer, and they improve every day against evals. Disruption is happening today faster than ever before because of AI.
How agentic intelligence products disrupt legacy software
Adapted from Clay Christensen’s model of disruptive innovation
Legacy software organizes the inputs
Records, workflows, and dashboards help the customer manage the job. The customer still executes it.
Agents get the job done
An agentic product executes the job steps and is measured by how fast and accurately the job gets done.
The business model changes
Agents are priced based on the economic value created for customers. Seat counts stop being the unit of value.
Christensen found that incumbents survive disruption by building the new jobs-to-be-done business in an autonomous team. This is why we built thrvAI as our agentic platform for equity value creation.
Learn more about our JTBD methodOur process
How we start or buy a portfolio company.
Step 01
We identify markets that meet our criteria
We use thrvAI to map the customer’s job-to-be-done, identify the unmet needs and competitor weaknesses, and size the market using the customers' willingness to pay. We are running this analysis on hundreds of markets.
Step 02
We build v1 of the new agentic product
When a market meets our criteria, we build v1 of the new agentic intelligence product with thrvAI to validate the customer value and reduce innovation risk.
Step 03
We iterate to product-market fit
We put the product in the hands of customers and iterate rapidly with thrvAI agents until the new product reaches product-market fit.
Step 04
We start a new company or acquire an existing one
Once our new product has product-market fit, we either fund a new company with Series A capital or we acquire an existing company to go to market. In both cases we use thrvAI to go to market and to continue building the new agentic product.
Disruption teams have launched some of the most successful products in history.
Apple iPhone2004 to 2007
Steve Jobs pulled a small team out of the iPod and Mac groups and moved them into a locked floor at 2 Infinite Loop, badge-restricted, windows papered over, with a sign reading “Fight Club” on the door. Members could not tell spouses or colleagues what they were working on. Two internal teams competed for the design: Tony Fadell’s group scaling up the iPod OS, and Scott Forstall’s group shrinking OS X. Forstall’s won. The isolation let Apple abandon the existing product’s interaction model instead of extending it, which is what a normal roadmap process would have forced.
Google Gmail2001 to 2004
Paul Buchheit started in August 2001 with an official charge to build an email product, and had a working prototype in a day by reusing search code from Google Groups. The project ran nearly three years with a team that reached about a dozen people at launch. Skeptical colleagues attacked it on technical and philosophical grounds, and the heavy use of JavaScript was widely considered a mistake, since browsers of the period might not survive it. Launch was never guaranteed. By early 2004 almost everyone at Google was using it for internal mail. It launched publicly on April 1, 2004, with a gigabyte of storage, hundreds of times what Hotmail and Yahoo offered.
Google Chrome2006 to 2008
Eric Schmidt opposed building an independent browser for six years, on the grounds that Google was a small company and he did not want a browser war. Larry Page and Sergey Brin hired several Mozilla Firefox developers and had them build a demonstration, after which Schmidt changed his position. Ben Goodger and Darin Fisher led the effort from 2006, and the team concluded that only a ground-up redesign would work. Isolating each tab in its own sandboxed process was the central idea, a departure from the single-process model every existing browser used. The first beta shipped on September 2, 2008. As of March 2026, StatCounter puts Chrome’s desktop share at just over 75%.
Amazon Kindle2004 to 2007
Jeff Bezos set up a separate subsidiary in Silicon Valley under Gregg Zehr, formerly of Palm, with its own hiring, its own name, and no obligation to look like Amazon. Amazon was a retailer with no hardware competence and every internal incentive to protect print margins. A wholly separate entity was the only way to build a device whose explicit purpose was to cannibalize the company’s largest category. The first units sold out in five and a half hours.
Sony PlayStation1992 to 1994
Ken Kutaragi’s sound chip work for Nintendo’s SNES was done semi-covertly, and Sony’s consumer electronics leadership wanted the whole games effort killed after Nintendo publicly walked away from the joint CD-ROM deal at CES in 1991. Norio Ohga’s solution was to move Kutaragi and his engineers out of the electronics division entirely and into Sony Music Entertainment Japan, a separate company with its own P&L, no hardware orthodoxy, and existing competence in manufacturing and distributing CDs to retail. That relocation became Sony Computer Entertainment in 1993.
Microsoft Internet Explorer1994 to 1995
Ben Slivka created the browser project inside the Windows 95 team in early October 1994 and staffed it with six people. Thomas Reardon licensed the Mosaic codebase from Spyglass in December, and the team mapped out an eight-week development plan in January. Microsoft’s official internet strategy at the time was MSN, a proprietary online service in the mould of CompuServe, and Slivka and J. Allard had spent the year arguing internally that the company should ride the open web instead of competing with it. Internet Explorer 1.0 shipped in August 1995 as part of the Internet Jumpstart Kit in Microsoft Plus.
Tencent WeChat2010 to 2011
Allen Zhang formed a team of ten in Guangzhou, an hour and a half by road from Tencent’s head office, and ran it in direct competition with another internal team building a mobile messaging app. Pony Ma allowed the two units to race each other. The new product competed head-on with Mobile QQ, then the cornerstone of Tencent’s social and gaming business, and a free messaging service also threatened the carriers’ SMS revenue. Seven engineers built the first version in about three months. It launched on January 21, 2011, and crossed 100 million users 433 days later. Instagram took 26 months to reach the same mark, Facebook took 52 months.
IBM Personal Computer1980 to 1981
Don Estridge got twelve engineers in Boca Raton, roughly 1,200 miles from Armonk, with authority to bypass IBM’s normal development and procurement rules. That exemption is what produced the decisions IBM is remembered for: off-the-shelf Intel processor, licensed DOS from Microsoft, published bus specification, retail distribution through ComputerLand and Sears. The machine shipped about a year after kickoff, against IBM’s typical four to five year cycle, and defined the industry standard.
An agentic product is a different business.
It needs a
different team.
New product, new platform, new pricing.
Legacy software organizes the inputs to your customer's job-to-be-done, but the customer still has to execute their job manually. Our new agentic products get the job done for customers. Selling and pricing an agentic product is based on the economic value created for customers instead of features and seats. With thrvAI, we ship agentic intelligence product improvements for our portfolio companies daily against evals. Quarterly feature releases become a relic.
New agentic products are the key to growth
We build new products with thrvAI using disruption teams. Disruption teams are a product innovation best practice with a proven track record of success. And they are more important now than ever before because a new agentic product requires a different innovation process, operating model, business model, and go to market plan. The new team has to be free from the incentives, metrics, and processes that worked for legacy products.
Everyone is a venture investor now.
All software equity value created today is with new agentic intelligence products that get the customer's job done faster and more accurately than legacy products. But most new products fail. We use thrvAI to reduce risk by taking multiple shots on goal like a venture investor. Innovation is a risk mitigation process, and the customer's job-to-be-done enables us to mitigate risk at the earliest stages of new product development.
What we build for each company
Full self-driving agents for the customer’s job
Our JTBD harness drives the whole job to completion. It maintains the state of the entire job, ingests updates, runs agents concurrently, and determines the next action. thrvAI creates the intelligence that guides the agents to refine their work and adjust the plan as they learn, carrying the job from request to result without the customer managing every step.
thrvAI maps the customer’s job and builds specialized agents to Plan, Execute, Assess, Revise and Conclude the job. Each agent owns a distinct part of the job, with the context, tools and expertise to get it done on the customer’s behalf.
thrvAI runs evals autonomously, continuously improving the economic value created for customers as measured by the metrics in the job.
When a run fails, thrvAI reads the trace, labels the failure, finds the first span where it went wrong, and fixes the error.
Each product gets an MCP tool surface, so its job agents can read and act on any of the customer’s systems of record.
We train proprietary open-weight models for each job step on the company’s own job executions and expert preferences.
Customers pay for the outcome of getting the job done, and the cost of each job execution is managed as a gross margin line.
Market opportunity
thrvAI uses the customers’ JTBD to identify the market opportunity for agentic growth.
Traditional market definitions are based on products, and the size of the market is based on the price of the product and the number of customers. To succeed at disruption, we define the market based on the customer’s job-to-be-done and their willingness to pay to get their job done. Then our thrvAI agents identify the steps, needs, metrics, root causes, and constraints in the customer’s JTBD. Our thrvAI agents analyze the competitors’ weaknesses and create the optimal agentic product strategy in the target market.
JTBD Harness
thrvAI builds a JTBD agent harness that drives the job successfully.
thrvAI creates a JTBD Agent Harness that drives the whole job to completion for the customers. The harness maintains the state of the entire job, ingests updates, runs agents concurrently, and determines the next action to get the job done. It's intelligence guides the agents to refine their work and adjust the plan as they learn, carrying the job from request to result without the customer having to manage every step.
Legacy business
Roadmap commitments and sprint planning decide what ships next.
Agentic business
thrvAI uses the customer’s job to decide what each agent does, in parallel.
Job Step Agents
thrvAI builds an autonomous agent team for the customer’s job.
thrvAI maps the customer’s job-to-be-done and builds specialized agents to Plan, Execute, Assess, Revise, and Conclude the job for the customer. Each agent owns a distinct part of the job, with the context, tools, and expertise to get the job done on the customer’s behalf.
Legacy business
Feature specs and acceptance criteria describe how screens should behave.
Agentic business
thrvAI creates job step agents that create economic value for the customer.
Evals
thrvAI creates evals that improve accuracy every day.
thrvAI runs evals autonomously in the background. Domain experts label the cases that define a job done correctly, including the edge cases only experienced practitioners recognize, and thrvAI generates new cases from every job execution. It runs the suite on every change, scores accuracy, and ships the improvements that raise it.
The team sets the standard for a job done correctly. thrvAI does the work of meeting it, around the clock, as it builds the agentic intelligence product that gets the customer’s job done for them.
Legacy business
QA tests features against expected behavior before a release train leaves.
Agentic business
thrvAI runs evals autonomously on every change, and improvements ship daily.
Error analysis
thrvAI conducts error analysis on every failed run.
thrvAI conducts error analysis autonomously in the background. When a run fails, thrvAI reads the trace, labels the failure, finds the first span where it went wrong, and updates the failure taxonomy. It then fixes the cause in the harness, the evals, or the model, re-runs the evals, and ships the fix when accuracy improves.
The team reviews what changed. The product that gets the customer’s job done keeps improving without waiting for a release.
Legacy business
Support tickets arrive after the customer has already been let down.
Agentic business
thrvAI finds, labels, and fixes failed runs before they become tickets.
MCP tool surface
thrvAI generates an MCP tool surface that works on any system of record.
thrvAI exposes the new product’s capabilities through an MCP tool surface, so the agents can read and act on other vendors’ systems of record. Customers can put the agents to work without replacing the software they already run.
The market grows to every customer who has the job, including customers running a competitor’s system.
Legacy business
The product creates value only inside its own database.
Agentic business
thrvAI agents get the job done using the customer’s data wherever it lives.
Proprietary models
thrvAI trains custom models on the customers’ job executions.
Every job execution and every expert correction is training data. thrvAI trains open-weight models for each job step. Each step gets a model that is more accurate for that step and less expensive to run than a general-purpose model.
The more jobs the new product executes, the more accurate and less expensive each model becomes.
Legacy business
Story points and feature delivery measure progress.
Agentic business
thrvAI creates open-weight models based on the job execution to create economic value for the customer at lower cost.
Pricing and margin
thrvAI creates a new agentic product priced on the economic value created for the customer.
Agents are priced against the economic value of the job, not the number of seats. thrvAI supports outcome or consumption pricing and tracks the cost of each job execution (inference, tool calls, and expert escalation) as a gross margin line.
Customers pay for jobs done. We manage the margin on each one.
Legacy business
Seat-based pricing and renewal management.
Agentic business
thrvAI creates a new agentic product priced on the economic value created for customers with cost per job execution managed as gross margin.
Why we do it
Intelligent agents will get the job done for customers to create economic value and improve their lives.
This is the reason behind each of our investment criteria. The most important is human impact: helping customers get their jobs done better so they can live better lives.
Learn more about our thrv Portfolio.
If you are an investor, an executive, or an expert in a complex customer job, get in touch. We would like to meet you.